n-channel mosfet rise-fall time comparison

[Jamie Morken]

Hi,

This 100V n-channel mosfet:
http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail?name=IPP04CN10NGIN-ND

has a specified rise time about 3x slower than the fall time,
(78ns rise time and 25ns fall time with 1.6ohm gate resistor)

and this 75V n-channel mosfet:
http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail?name=IRFB3077PBF-ND

has a specified fall time about equal to the rise time.
(87ns rise time and 95ns fall time with 2.1ohm gate resistor)

What are some of the factors inside an n-mosfet that effect the mosfet
rise/fall times? Or are these results just because of the test setup
used? ie. perhaps schottkys parallel with the gate resistors were used
in the case of fast fall times.

Also for an H-bridge, can switching losses be minimized by taking these
relative switching speeds into account? For example by putting
different mosfets with faster fall times on either the upper or lower
legs of the Hbridge, assuming the same total switching times for the
upper and lower fets used.

cheers,
Jamie

 

[Jamie Morken]

My mental model of a mosfet is an infinitely fast chip that has
parasitic capacitance on-chip and wirebond inductance out to the
leads. In the practical case, most mosfets will switch much faster
than the datasheets suggest if you can drive the gates hard enough and
keep external circuit parasitics down. The rise/fall difference on the
datasheets is mostly an artifact of the test circuit: turn-on
impedance (Rds-on) is a lot lower than turn-off (whatever else is out
there.)

This is a 50-volt pulse into a 50 ohm load, transformer coupled to
boot:

ftp://66.117.156.8/HV_mosfet_pulse.jpg

using a couple of 5-cent 2N7002's.

Hi,

Looks like under 5ns rise/fall times Do you have a schematic of the
circuit?

cheers,
Jamie

 

[Jamie Morken]

Sorry, can't do that. But the points are, I suppose,

use several small fets instead of one big one, to reduce wirebond
inductance

keep the layout tight

drive the gates really hard.

don't take the datasheets too seriously

Thanks, also I was interested in your transformer coupled design to
compare to the one I'm using

cheers,
Jamie

 

[Tim Williams]

What are some of the factors inside an n-mosfet that effect the mosfet
rise/fall times?

Affect.

What is Vgs(th)? I can imagine that, if V(low) isn't as far below
Vgs(th) as V(high) is above it, that will directly affect how fast the
miller capacitance charges through the gate resistor (and whatever
source impedance there is).

Re: high speed stuff- parallel stuff should come as no surprise
(although I am a little surprised 2N7002's are a bit faster than
2N3904s ). That's exactly what they do for RF stuff- you can get
chips that consist of myriad little junctions (or FET channels),
either strategically wired together with metallization or tediously
wired, individually, with bond wires! Think of a wafer of 2N3904
dies, without cutting them apart, and lashing them all together:

100MHz fT in the size and ratings of a TIP42. It's no wonder RF

transistors are pricey!

Tim

 

[Mike Monett]

On Sun, 14 Sep 2008 17:47:49 -0500, John Fields

So you're measuring it something like this:
Vin>---+
|
[49.5R]
|
+-----<<---+
| |
[0.5R] [SCOPE]
| |
GND>---+-----<<---+
or, maybe:
Vin>--+--<<--+
| |
| |
[50R] [1/100]-->>--[SCOPE]--+
| | |
| | |
GND>--+--<<--+----->>-----------+
JF

No, just a 40 dB, 18 GHz SMA attenuator. The scope is a Tel 11801
with a 20 GHz sampling head.

John

That still doesn't explain how you hooked it up.

The photo cleverly doesn't show the entire waveform, so you can't
tell if it's a positive or negative-going pulse, or what the duty
cycle is.

Assuming the gates are driven from a low impedance, that probably
means a pulse generator with 50 ohm source impedance, perhaps
driving an attenuator to further reduce the Thevenin impedance at
the gates. The case of the generator is at ground, so the 2N7002
sources would be at ground also.

So the only thing left to wiggle is the drains. And that is a
problem.

The attenuator is a 40 dB 18 GHz attenuator, so it has an input
impedance of 50 ohms.

There doesn't seem to be a coupling cap between the drain and the
attenuator, since the waveform shown goes from zero to +0.5V.

The attenuator case is grounded, so you are probably not connecting
the center pin of the attenuator to +50V and driving the outside
shield with the drain of the 2N7002. That would mean lifting the
scope off ground, which is difficult to do at 1GHz.

If you connected the input pin directly to the drains of the 2N7002,
you would need to connect a 50 ohm resistor to a 100V supply in
order to get a 50V swing at the drain.

But that would burn out the attenuator if the 2N7002's were left
off.

But with the 2N7002's turned on, the 50 ohm drain resistor would
dissipate 200 watts.

That would mean you are using a very good high frequency resistor,
since there is little or no overshoot in the waveform. That would be
very expensive.

You might be cheating a bit by switching the circuit on briefly to
capture the waveform, then turning it off to save the resistor. But
that seems unlikely.

The 2N7002's are only rated for 60V VDSS, so there's not much room
for other approaches that might involve higher supply voltages.

So it's not definite how you are getting the waveform, but it is
possible to eliminate many of the options due to bandwidth, cost, or
power considerations.

That leaves few remaining alternatives, so your circuit has to be in
one of them.

So what's your trick?

Best Regards,

Mike Monett

 

[Mike Monett]

Why is that a problem? It is, as I've said, transformer coupled.

So why was JF fooling with direct coupling. And why did you say you
couldn't post the schematic?

Anyway, I figured you had to be using a transformer while taking a
shower.

I got XNews to retrieve the whole thread, and found it. Should have
done that at the beginning

You invert the output signal to show a positive-going pulse. That's
what threw me off.

And a very low duty cycle to avoid burning out the attenuator.

Like I said: two 2N7002's driving a transmission-line isolation
transformer, with brutal gate drives. The point being that mosfets
can be made to switch a lot faster than their datasheets usually
suggest.

Next time, we're going to use a GaN fet. That should really
scream.

Nice. Please post the pictures so we can enjoy!

John

Mike Monett

 

[[email protected]]

My mental model of a mosfet is an infinitely fast chip that has
parasitic capacitance on-chip and wirebond inductance out to the
leads. In the practical case, most mosfets will switch much faster
than the datasheets suggest if you can drive the gates hard enough and
keep external circuit parasitics down. The rise/fall difference on the
datasheets is mostly an artifact of the test circuit: turn-on
impedance (Rds-on) is a lot lower than turn-off (whatever else is out
there.)

This is a 50-volt pulse into a 50 ohm load, transformer coupled to
boot:

ftp://66.117.156.8/HV_mosfet_pulse.jpg

using a couple of 5-cent 2N7002's.

John

There is distributed resistance in the gate. That is, the gate is
poly, but picked up well with "metal". [Metal is often a mixture of
stuff on chips, hence the quotes.]

I wasn't 100% sure of this since I've only done lateral flow devices,
but found a good paper on power semiconductors in general:
http://ewh.ieee.org/r8/ukri/pels/Mawby.pdf

 

[Mike Monett]

[...]

No, the sampling head has a 50 input impedance.

The 40 dB pad resistors are 51.01, 2499.75, and 51.01.

John, a series resistor of 2499.75 ohms might be impractical for
this attenuator.

A stray capacitance of only 3.53E-15 F. between the ends would have
the same reactance as the series resistor at 18Ghz. This would
probably put the unit out of spec.

Also, a small amount of capacitance from the center of a long high
value resistor to the outside shield might have serious effects on
the bandwidth. For example, see Win's description of "High voltage
probe frequency response", at

http://www.repairfaq.org/REPAIR/F_hvprobe.html#HVPROBE_008

However, if the attenuator were split into three cascaded 10dB
sections, the resistors would be 96.25, 71.15, and 96.25 ohms. Ref:

http://www.random-science-tools.com/electronics/PI_attenuator.html

This would dramatically reduce the effects of stray capacitance.

This is getting tiresome.

Are you sure you are considering possibilities?

I find it stimulating, probably because I like to consider
possibilities

John

Mike Monett

 

[Mike Monett]

[...]

No, the sampling head has a 50 input impedance.

The 40 dB pad resistors are 51.01, 2499.75, and 51.01.

John, a series resistor of 2499.75 ohms might be impractical for
this attenuator.

A stray capacitance of only 3.53E-15 F. between the ends would have
the same reactance as the series resistor at 18Ghz. This would
probably put the unit out of spec.

Also, a small amount of capacitance from the center of a long high
value resistor to the outside shield might have serious effects on
the bandwidth. For example, see Win's description of "High voltage
probe frequency response", at

http://www.repairfaq.org/REPAIR/F_hvprobe.html#HVPROBE_008

However, if the attenuator were split into three cascaded 10dB
sections, the resistors would be 96.25, 71.15, and 96.25 ohms. Ref:

http://www.random-science-tools.com/electronics/PI_attenuator.html

This would dramatically reduce the effects of stray capacitance.

This is getting tiresome.

Are you sure you are considering possibilities?

I buy attenuators from Mini-Circuits. I don't know what's inside, but
the black-box equivalent is the pi network shown. I have seen
attenuators and terminators that are thinfilm sheets, not lumped
components at all.

You have discussed making a simple 20dB resistive attenuator probe for low
microwave frequencies in the past. I believe you recommended using a
Caddock resistor for the series element. This would work fine at low
microwave frequencies.

A thin film sheet might be also suitable for these frequencies, perhaps up
to 5GHz or so. But have you see it used in a 40dB attenuator for 18GHz?

The series resistor would have low stray capacity between the ends, but it
would still have capacity to the shield. Win's analysis shows this might
give problems at 18GHz. Also, depending on the length, the stray inductance
might become a problem at 18GHz.

So you want the length to be as short as possible to reduce the inductance,
but this increases the capacitance betwen the ends.

Alltogether, high values of resistance are difficult to use at these
frequencies.

And I might note that 3 times 10 dB is not 40 dB.

OK, put four 10dB sections in cascade. The resistors are cheap, and
the values remain the same.

John

Mike Monett

 

[Mike Monett]

[...]

Darn, now you've got me curious. I wish people wouldn't do that.

Why? How can you consider possibilities without curiosity?

ftp://66.117.156.8/VAT-20.zip

That's a Mini-Circuits VAT-20, a cheap 6 GHz, 20 dB attenuator.

The 18 GHz parts are a lot smaller.

What's the schematic look like? The black rectangles look like thick
film resistors. But you can barely see the outline of something else
on the surface. Is that thin-film?

Also, it looks like a single section 40dB attenuator for 18GHz may
be difficult for a number of other reasons besides the ones I gave
above. I did some research on microwave attenuators.

Agilent states their attenuators use thin film:

"Agilent attenuators achieve flat-frequency response and high
accuracy through the use of thin-film attenuator cards. These cards
are composed of high-stability tantalum nitride resistive film,
deposited on sapphire or alumina substrates."

http://www.nit.eu/oferta/wypozyczanie/pdf/g30/Agilent-8494B.pdf

Vishay sells microwave thin-film resistors that go up to 2kohm:

"Industry's Smallest Thin-Film Resistors for Microwave Applications
Offer High Performance in Chip as Small as 0.010" by 0.020""

http://www.vishay.com/company/press/releases/2003/030425resistor/

Hittite recently announced three wideband attenuators for the DC to
25 GHz frequency band. The maximum attenuation is 20dB:

http://www.hittite.com/products/view.html/view/HMC656LP2

Barrie Industries offers a thin-film chip attenuator. The maximum
attenuation is 30dB:

http://www.barryind.com/en/AT.html

IMS offers thin-film attenuators for DC to 20 GHz. The maximum
attenuation is only 10dB. The higher attenuations have a much lower
frequency range:

http://www.ims-resistors.com/A-series.pdf.

MSI offers thin film attenuators to 24dB at 20GHz:

http://www.rikei.co.jp/dbdata/pdf/MSAT567.PDF

So the maxium attenuation I could find in a single section
attenuator for 20GHz is 30dB or less. If a 40dB attenuator exists,
it didn't show up in an extensive search.

I did find another on-line attenuator calculator. It handles PI, T,
Bridged-T, and Balanced attenuators. It doesn't require javascript:

http://www.microwaves101.com/encyclopedia/calcattenuator.cfm

Benjamin Lewis has calculators that allow different input and output
impedances. They also don't require javascript:

http://benl.co.uk/webapps/attenuator-calculators/pi-type

Microwaves 101 has an excellent page on microwave thin-films. They
explain that 2 microinches of tantalum nitride (TaN) has a DC
resistance of about 50 ohms per square. They go on to explain:

"This thickness of TaN is less than 1% of a skin depth at X-band, so
the RF sheet resistance is very nearly equal to the DC value. The
plot below shows how the RF skin depth varies over frequency; the
error is only about 1% all the way up at W-band, less at lower
frequencies. Nothing to concern yourself with."

http://www.microwaves101.com/encyclopedia/RF_sheet_res_examples.cfm

So thin film is a necessity at these frequencies due to skin effect.

Here's the catch of the day. Triquint Semiconductors show how to
calculate the attenuation of the evanescent wave in microwave
cavities. The following paragraphs are especially interesting:

"Design Guidelines for Microwave Cavities"

"These equations provide a good first order approximation to the
problem and can sometimes highlight serious radiation issues before
the design is frozen. Due to the many other variables which can add
to, or subtract from, the radiating or propagating signal
(bondwires, substrates, microwave structures, filters, passive
components, etc), it is best to stay as conservative as the design
will allow. At higher frequencies such as 40 GHz it becomes
difficult to build a channel below the waveguide cutoff frequency
(b=0.147 inches at 40 GHz) and still support the circuit element
sizes. To achieve a 3x ratio at 40 GHz would require a channel width
of 0.049 inches and a height from module floor to lid less than this
value."

"If the design dictates that active components, such as MMIC
amplifiers, be placed in a propagating waveguide channel, it is
prudent to limit their gain to 20 or 30 dB maximum. The use of
absorber on the lid in this case will almost always be required and
some gain ripple due to radiative feedback of the output signal can
be expected. The best course of action is to keep everything very
close to the ground plane. This reduces to a minimum the radiation
of components such as bondwires and other transitions. It is not
uncommon for a MMIC amplifier with 15 or 20 dB of gain to lose about
1 dB when a lid with absorber is placed above the MMIC. This is an
indication that the radiative signal level is not negligible."

http://www.triquint.com/prodserv/tech_info/docs/mmw_appnotes/DesignGuidel
inesApNote.pdf

This indicates you might want to stay at or below 20 dB in
single-section precision attenuators, as well as wideband
amplifiers.

For critical high frequency work, you might want to look at chip
bonding. Luis Cupido has started a Yahoo group to discuss the
issues. The registration is free.

http://tech.groups.yahoo.com/group/chip-hybrids/

(Thanks to John Miles KE5FX for mentioning this group in one of the
mailing lists.)

Luis Cupido's web site is a cornucopia of information for anyone
interested in high frequency work. He even describes a Corner Cube
Harmonic Mixer for 411GHz:

http://w3ref.cfn.ist.utl.pt/cupido/

We are surrounded by a wealth of information that is available
instantly for little or no effort. But the answers to one question
immediately create more questions.

How can you not be curious?

John

Mike Monett

 

[Mike Monett]

[...]

Darn, now you've got me curious. I wish people wouldn't do that.

Why? How can you consider possibilities without curiosity?

ftp://66.117.156.8/VAT-20.zip

That's a Mini-Circuits VAT-20, a cheap 6 GHz, 20 dB attenuator.

The 18 GHz parts are a lot smaller.

What's the schematic look like? The black rectangles look like thick
film resistors. But you can barely see the outline of something else
on the surface. Is that thin-film?

Thickfilm on alumina; screened and fired conductors, screened and
fired cermet resistor elements, laser trimmed.



in >-----+-----R2-------R3------+------< out
| |
| |
R1 R4
| |
| |
| |
gnd>-----+----------------------+------< gnd

The R2-R3 node is a pretty big hunk of conductor, essentially a C to
ground, probably to kill capacitive shoot-through, maybe to
deliberately limit the bw to 6 GHz.

A la Winfield's high voltage probe. But the thick film will also have a
built-in frequency limit due to skin effect, as described in the Microwaves
101 web page.

You said the 18 GHz parts are smaller. What do they look like?

John

Mike Monett

 

[Mike Monett]

On Tue, 16 Sep 2008 05:54:36 +0000, Mike Monett

[...]

Darn, now you've got me curious. I wish people wouldn't do that.

Why? How can you consider possibilities without curiosity?

ftp://66.117.156.8/VAT-20.zip

That's a Mini-Circuits VAT-20, a cheap 6 GHz, 20 dB attenuator.

The 18 GHz parts are a lot smaller.

What's the schematic look like? The black rectangles look like
thick film resistors. But you can barely see the outline of
something else on the surface. Is that thin-film?

Thickfilm on alumina; screened and fired conductors, screened and
fired cermet resistor elements, laser trimmed.



in >-----+-----R2-------R3------+------< out
| |
| |
R1 R4
| |
| |
| |
gnd>-----+----------------------+------< gnd

The R2-R3 node is a pretty big hunk of conductor, essentially a C to
ground, probably to kill capacitive shoot-through, maybe to
deliberately limit the bw to 6 GHz.

A la Winfield's high voltage probe. But the thick film will also have
a built-in frequency limit due to skin effect, as described in the
Microwaves 101 web page.

You said the 18 GHz parts are smaller. What do they look like?

Ask Mr Google.

John

Thanks, John. That's money in the bank.

Mike Monett

 

[[email protected]]

Hi,

This 100V n-channel mosfet:http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail?name=IPP0...

has a specified rise time about 3x slower than the fall time,
(78ns rise time and 25ns fall time with 1.6ohm gate resistor)

and this 75V n-channel mosfet:http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail?name=IRFB...

has a specified fall time about equal to the rise time.
(87ns rise time and 95ns fall time with 2.1ohm gate resistor)

What are some of the factors inside an n-mosfet that effect the mosfet
rise/fall times?  Or are these results just because of the test setup
used?  ie. perhaps schottkys parallel with the gate resistors were used
in the case of fast fall times.

Also for an H-bridge, can switching losses be minimized by taking these
relative switching speeds into account?  For example by putting
different mosfets with faster fall times on either the upper or lower
legs of the Hbridge, assuming the same total switching times for the
upper and lower fets used.

cheers,
Jamie

Not sure what your up to with this Jamie, but here is a tip. After you
finish your circuit don't discount the fact that the drain voltage can
oscillate in production models. Just to pass this on we had to use a
snubber on a simple switching amp to damp oscillations. There was no
reason that we could see for this to happen but it did. So we loaded
the drain with a series resistor and cap to ground and dumped the
unwanted frequencies. They were are arcing the boards like an rf burn.
A very tiny hole straight out of the body of a resistor we used for
feedback that was directly connect to drain. You can tailor the rise
and fall times of the fet by sourcing and sinking the gate drive with
different impedance on each driver.

 

[[email protected]]

Hi,

This 100V n-channel mosfet:http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail?name=IPP0...

has a specified rise time about 3x slower than the fall time,
(78ns rise time and 25ns fall time with 1.6ohm gate resistor)

and this 75V n-channel mosfet:http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail?name=IRFB...

has a specified fall time about equal to the rise time.
(87ns rise time and 95ns fall time with 2.1ohm gate resistor)

What are some of the factors inside an n-mosfet that effect the mosfet
rise/fall times?  Or are these results just because of the test setup
used?  ie. perhaps schottkys parallel with the gate resistors were used
in the case of fast fall times.

Also for an H-bridge, can switching losses be minimized by taking these
relative switching speeds into account?  For example by putting
different mosfets with faster fall times on either the upper or lower
legs of the Hbridge, assuming the same total switching times for the
upper and lower fets used.

cheers,
Jamie

Jamie, so what, who says you have to drive the fet with a single gate
resistor. Use a current sink and a current source with independent
impedance to control the fet timing.